OpenAI releases GPT-5.1

OpenAI released GPT-5.1, an update to GPT-5 released three months earlier, in Instant and Thinking variants alongside a Pro tier and a compact coding-focused model, GPT-5.1-Codex-Mini. OpenAI described the changes as being about tone and behaviour as much as raw capability: the model was tuned to be “warmer” by default, and users could choose among eight preset personalities, from Professional to Candid to Quirky, to change how it phrased responses.

The more substantive change was to GPT-5.1 Instant, which for the first time could decide on its own to pause and reason before answering a difficult question — such as a complex maths problem or a tricky bug — while continuing to answer simple queries immediately, rather than requiring the user to switch to a separate “Thinking” mode. API users retained finer control, with a reasoning-effort setting running from minimal to high. OpenAI reported improvements on math and coding evaluations including AIME 2025 and Codeforces over GPT-5, without publishing a head-to-head comparison against competing frontier models in the release material itself.

Early reviews were mixed: outlets praised gains in coding and mathematics and the more natural conversational tone, while others found the update incremental against rivals such as Google’s Gemini 3, released six days later, and against GPT-5 itself, framing it as a refinement of an existing model rather than a new generation. The eight-personality system drew particular attention as a response to earlier criticism that GPT-5’s default tone had been received as cold or overly terse compared with GPT-4o. A further variant, GPT-5.1-Codex-Max, aimed at long-running coding tasks, followed a week later.